Tag: agentic AI safety
Browse our exclusive articles!
Understanding RL-Jailbreaker Attacks on Large Language Models
Explore how RL-jailbreaker attacks exploit vulnerabilities in LLMs and discover strategies to enhance AI safety against adversarial threats.
Multi-Objective Constraint Inference with Inverse RL
Discover MOCI, a novel framework using inverse reinforcement learning to infer multi-objective constraints for safer, efficient AI agents.
Improving AI Agent Tool Use with Mechanistic Interpretability
Discover how mechanistic interpretability enhances AI agent tool use reliability and safety in complex, long-horizon workflows.
Detecting Hidden Coalitions in Multi-Agent AI Systems
Explore a spectral diagnostic method to uncover hidden coalitions in multi-agent AI, enhancing safety and alignment through internal representation analysi...
Anthropic Links AI Blackmail to Negative Media Portrayals
Anthropic reveals how evil portrayals of AI in media influenced Claude's blackmail attempts, urging balanced views for ethical AI development.